Abstract: There is provided a method that includes obtaining (a) a dense representation of external features, (b) a dense representation of text, and (c) a mask that associates the external features to tokens of the text, and employing a cross-attention process that utilizes the mask to perform an information fusion of the dense representation of the external features and the tokens of the text, thus yielding a joint representation of the external features and the tokens of the text. There is also provided a system that executes the method, and a storage device that includes instructions for controlling a processor to perform the method.

Abstract: Using indentation to trim output of a language synthesis model. A computer system determines a first indentation level associated with a cursor within an editor buffer. The computer system identifies a block of line(s) from an output from a language synthesis model. Each line in the block meets at least one criterion from a set of criteria, including a first criterion that the line is an initial line in the output from the language synthesis model, a second criterion that the line has a second indentation level that is greater than the first indentation level, and a third criterion that the line contains a continuation token. The computer system identifies a particular line, immediately following the block, that does not meet any criterion from the set of criteria. The computer system trims the output from the language synthesis model from the particular line onward.

Abstract: Techniques to temporally filter images via a filtering weight computation are disclosed. A first image having a first timestamp and a second image having a second timestamp are acquired. These images are generated by a camera, and the first timestamp is before the second timestamp. A motion compensation (MC) operation is performed on the first image to produce an MC image. A difference image is generated using the MC image and the second image. The difference image reflects differences in intensities that exist between the two images. A local weight map is generated based on those differences. A global weight map is generated based on certain IMU data. A final weight map is generated based on a combination of the local weight map and the global weight map. The final weight map is used to generate a filtered image.

Abstract: Methods and computing devices for estimating a force F exerted on a touchpad are disclosed. In one example, a method comprises determining that the touchpad is not being touched. At least on condition of determining that the touchpad is not being touched, a no-touch capacitance value of the PCB is calculated. After calculating the no-touch capacitance value, the method includes determining that the touchpad is being touched. At least on condition that the touchpad is being touched, the no-touch capacitance value and a touch-based capacitance value are used to estimate the force F exerted on the touchpad.

Abstract: A reference to a digital item is stored as a digital card. The digital card can also be contained in, and/or refer to, other digital cards. The digital card can also include properties or attributes that may be added from the digital item that is being referred to. The digital card can be stored in a data pod within a de-centralized data storage system architecture.

Abstract: Computer-implemented systems and methods for rendering a visual effect on a graphical element, including: generating an initial visual effect based on the graphical element; computing a mask shape based on the graphical element; generating a mask based on the mask shape from the graphical element; masking the initial visual effect based on the mask to generate a masked visual effect; and combining the masked visual effect with the graphical element into a combined graphical element and visual effect.

Abstract: Compressing memory addresses within an execution trace via reference to a translation lookaside buffer (TLB) entry. A microprocessor identifies a TLB entry within a TLB slot, the TLB entry mapping a virtual memory page to a physical memory page. The microprocessor initiates logging of the TLB entry by initiating logging of at least a virtual address of the virtual memory page, and an identifier that uniquely identifies the TLB entry from among a plurality of live TLB entries. Subsequently, the microprocessor identifies a cache entry within a memory cache slot, the cache entry comprising a physical memory address corresponding to a cache line. The microprocessor initiates logging of the cache entry by matching a physical memory page identification portion of the physical memory address with the TLB entry, and then initiates logging of at least the identifier for the TLB entry and an offset portion.

Abstract: Embodiments of the present disclosure includes techniques for cache memory replacement in a processing unit. A first data production operation to store first data to a first cache line of the cache memory is detected at a first time. A retention status of the first cache line is updated to a first retention level as a result of the first data production operation. Protection against displacement of the first data in the first cache line is increased based on the first retention level. A first data consumption operation retrieving the first data from the first cache line is detected at a second time after the first time. The retention status of the first cache line is updated to a second retention level as a result of the first data consumption operation, the second retention level being a lower level of retention than the first retention level.

Abstract: A method of manufacturing is provided, including forming a metal frame of a head-mounted computing device shaped as a pair of eyeglasses. Forming the metal frame may include additively manufacturing the metal frame of the head-mounted computing device by performing laser sintering on aluminum powder or titanium powder. Forming the metal frame may further include removing an outer surface of the metal frame via a reductive process.

Abstract: Systems and methods are provided for obtaining, training, and using an end-to-end AST model based on a neural transducer, the end-to-end AST model comprising at least (i) an acoustic encoder which is configured to receive and encode audio data, (ii) a prediction network which is integrated in a parallel model architecture with the acoustic encoder in the end-to-end AST model, and (iii) a joint layer which is integrated in series with the acoustic encoder and prediction network. The end-to-end AST model is configured to generate a transcription in the second language of input audio data in the first language such that the acoustic encoder learns a plurality of temporal processing paths.

Abstract: Various embodiments provide a so-called companion experience in which content consumed on a primary screen can serve as a source for an automatic search that returns related content that can be presented on an auxiliary screen. The companion experience can be considered to reside in a layer that can be moved across different screens. The different screens can include different physical screens, such as those associated with different computing devices, or the same physical screen in which the companion experience would be rendered in a frame or sub-window.

Abstract: The present disclosure relates to systems, methods, and computer-readable media to selectively enhance digital image and video content. For example, systems disclosed herein can encode original video content to compress and decompress the original video content. Systems described herein can further identify area of interest information for use in identifying portions of decompressed video content to analyze and remove one or more compression artifacts found therein. Systems described herein can further enhance the decompressed video content by increasing resolution for display. By identifying areas of interest and selectively enhancing digital video content, the systems described herein can reduce consumption of bandwidth and processing resources while maintaining high visual quality of the digital content.

Abstract: In-place recovery of fatal system errors at virtualization hosts. A device identifies an occurrence of a fatal system error in the first instance of a host operating system (OS) executing in a computer system. The device determines to perform an in-place recovery for the fatal system error. The device performs the in-place recovery, including pausing the execution of a virtual machine (VM) by the first instance of the host OS, preserving a state of the VM within system memory of the computer system, and resuming the execution of the VM by a second instance of the host OS executing in the computer system based on the state of the VM that is preserved within the system memory of the computer system.

Abstract: A fabric control protocol (FCP) is a data transmission protocol that enables spraying of individual packets for a given packet flow across a data center from an ingress interface of the source data processing unit (DPU) across a plurality of parallel data paths of a logical tunnel in the network fabric to the egress interface of the destination DPU. The FCP has congestion control mechanisms used to determine a degree of congestion at the egress interface of the destination DPU and to modify a send window size at the source DPU based on the degree of congestion. Reliable FCP (rFCP) extensions provide reliability enhancements and improved failure resilience within the data center. The rFCP extensions provide an unsolicited mode for low latency operation with enhanced reliability mechanisms. The rFCP extensions provide failure resilience mechanisms to identify and avoid failed paths among multiple parallel data paths within the logical tunnel.

Abstract: Embodiments of the disclosed technologies are capable of providing engagement feedback for an online system while maintaining user privacy. Embodiments determine one or more user groups. A user group includes users, and a size of the user group is selected to protect user privacy. A user action is received from a user in a first group. The user action includes an interaction with a piece of content. A group action count is updated for the first group based on the user action. The group action count indicates a number of interactions with the piece of content by any of the users in the first group. An estimated number of unique engagements is calculated based on the group action count and the size of the first group. The estimated number of unique engagements is provided to the online system as the engagement feedback for the piece of content.

Abstract: One example provides a display device comprising a display panel, a touch sensor, an ambient light sensor, a processor, and storage. The storage comprises instructions executable by the processor to monitor an ambient light signal received from the ambient light sensor, detect, in combination with a call event, a threshold change in the ambient light signal that is indicative of the display device being proximate to a body of a user, and based at least on detecting the threshold change in the ambient light signal, modify an operation of the display device.

Abstract: The disclosed techniques provide enhanced eye tracking systems utilizing joint estimation of biological parameters and hardware parameters. A system uses joint estimation of biological parameters, e.g., direction and position of an eye, with concurrent estimation of hardware parameters, e.g., camera position or camera direction, to self-calibrate and provide eye tracking estimations to accommodate for deformations and other changes of a device. Sensor data is used to select hardware parameters of a camera for use in the joint estimation with the biological parameters, where the hardware parameters are estimated based on glint and pupil position of a user. The disclosed techniques include a method to model changes of a device, as well as detect and compensate for them while the eye-tracking device is in normal use, without requiring a factory-calibration procedure to be repeated.

Abstract: Examples are disclosed relating to computing devices and methods for performing touch detection within a virtual trackpad area of a touch screen display. In one example, a non-trackpad touch input signal is received from outside the virtual trackpad area and processed with at least a jitter restrictor algorithm that applies a non-trackpad distance between reported touch locations. A virtual trackpad touch input signal is received from within the virtual trackpad area. On condition of determining that the virtual trackpad touch input signal is received from within the virtual trackpad area, the virtual trackpad touch input signal is processed with the jitter restrictor algorithm that applies a virtual trackpad distance between reported touch locations that is smaller than the non-trackpad distance between reported touch locations.

Abstract: A computing system for preventing at least a portion of an input device from attaching to an improper location on a computing device is disclosed. In one example, the input device comprises first and second input device magnets spaced by a separation distance and having a first magnetic pole orientation. The computing device comprises a housing with a first side and a first end adjacent to the first side, and first and second computing device magnets spaced by the separation distance and having a second magnetic pole orientation opposite to the first magnetic pole orientation. At least one repelling magnet having the first magnetic pole orientation is located between the first end of the housing and the second computing device magnet to repel the second input device magnet of the input device.

Abstract: A processing unit includes a first die and a second die with a microfluidic volume between the first die and the second die. At least one heat transfer structure couples the first die to the second die and is located in the microfluid volume. An electrochemical fluid is positioned in the microfluidic volume to provide electrochemical energy to at least one of the first die and the second die and receive heat from the first die and the second die.